摘要
为解决运动车辆检测中车辆数目统计精确度不高、实时性不强等问题,提出一种基于改进的Adaboost算法和帧间差分法的检测方法.采用汽车图像样本的Haar-like特征训练级联分类器,通过三帧差分法得到二值化掩码图像,去除干扰噪声后寻找连通域、重建前景掩码图;加载分类器在当前帧的前景区域中做多尺度检测,标记检测出来的运动物体,统计运动车辆数目.实验结果表明:该方法可以有效对全图信息进行筛选剔除,减小信息量,从而大幅度地提高检测精度,提升目标的检测速度;该方法对交通路口有效检测区域的车辆检测有很好的检测效果,对于路面复杂背景下的车辆依然有很高的检测率.
The methods of moving vehicles detection were normally difficult to count the correct numbers of the vehicles and they may not be able to reach the real-time requirement.To solve those problems,a vehicle detection algorithm was proposed based on improved Adaboost algorithm and frame differencing rule.A cascaded classifier was trained using Haar-like features of the vehicle image samples,while a binary mask image was created by the three-frame differencing rule,then a foreground mask image was rebuilt by removing interference noise and finding connected domains,finally the cascaded classifier was loaded to make a multi-scale detection in the current frame′s foreground region, and then the moving vehicles were marked and counted.The results of experiments show that this algorithm can effectively screen out the useful information,by this way,the detection accuracy and speed of the moving vehicles are highly improved,this algorithm is capable of detecting the moving vehicles in effective area of the traffic intersection.It was also induced a strong response under a complex motion.
出处
《华中科技大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2013年第S1期379-382,共4页
Journal of Huazhong University of Science and Technology(Natural Science Edition)
基金
湖北省教育厅科学研究计划重点资助项目(D20111509)
武汉工程大学研究生教育创新基金资助项目(CX201273)
